Inducing context-dependent decay with probe blocks

To quantify motor memory de-adaptation, we used probe blocks of 14 trials. The first and last two trials of each probe block were channel trials on the movement to the secondary target to

i

disappears mid-movement wait in middle of workspace channel

Fig. 4.1 Experimental paradigm to isolate how planning and execution of lead-in ments affects de-adaptation in different contexts. A) Participants made an initial move-ment from a starting location (bottom gray circle) to a central target (middle gray circle), and then immediately moved to a secondary target (yellow circle). On exposure trials, a velocity-dependent curl force field (force vectors shown as blue arrows for a typical straight line movement to the secondary target) was applied during this second movement, and the field direction, clockwise (CW) or counter-clockwise (CCW), was determined by the start location (either +45^◦southeast or -45^◦southwest of the primary target). Note that the sec-ondary movement was over the same space for both lead in and only shown separately here for clarity. B) In the probe and re-exposure phase, we tested how performing probe blocks containing different trial types affected de-adaptation of the two adapted motor memories.

Each probe block began and ended with a channel trial in each context (a left and right lead-in). We assessed de-adaptation under six different types of intervening probe trials (i-vi). Participants performed one of the six trial types for 8-10 trials and we tested the level of adaptation before and after each block to measure de-adaptation. Participants were then re-exposed to force-field perturbations to restore their adaptation level prior to the next de-adaptation block, in which another trial type was tested. In the planning-only probe blocks we included in the intervening trials two intermixed channel trials which were made in the same context, to encourage planning of full two-part movements. We also included these in the wait time probe blocks to control for their effect on net de-adaptation.

assess the level of adaptation for each of the two contexts (±45^◦lead-ins in a balanced order across repetitions). The intervening 10 trials consisted of one of 6 different probe types: i) 10 channel trials in the left context, ii) 10 channel trials in the right context, iii) 8 planning-only trials and 2 channel trials both in the left context, iv) 8 planning-only trials and 2 channel trials both in the right context, v) 8 trials in which subjects waited passively midway between the start locations and 2 channel trials in the left context vi) 8 trials in which subjects waited midway between the start locations and 2 channel trials in the right context (Figure 4.1).

The two channel trials were included so as to encourage planning of full movements in the planning-only probe blocks.

The wait probe blocks were included to contrast the effects of intervening trials and time, on de-adaptation. On wait ‘trials’ the subject’s hand position was maintained midway between the two start positions for the average duration of the previously executed trials. No targets or auditory cues were made between wait time trials. The two channel trials were included in wait probe blocks to match the planning-only probe blocks. These conditions allowed us to assess how de-adaptation depends on context (same or different) and on planning or execution.

Therefore, a set of intervening trials was either (primarily) channel trials, planning-only trials, or wait time trials, for each of the two start locations. On planning-only trials we increased the duration of the inter-trial interval so that the total duration of each planning-only or wait time trial was equal to the average duration of a full lead-in trial. This enabled us to isolate the effect of trial type from that of time.

Each probe block therefore consisted of a total of 43% channel trials, a similar ratio to the 33% channel trials we used previously in a follow-through task in which planning alone led to learning of opposing force fields (Sheahan et al., 2016).

After each probe block there was a short re-exposure period of between 26 and 32 exposure trials. We probed adaptation using 4 repetitions of each of these 6 conditions (336 de-adaptation trials plus 696 re-exposure trials) with each condition presented in a pseudorandom order between each rest break. Finally a post-exposure phase of 6 blocks (60 null trials) was performed.

4.2.3 Data analysis

Assessments of learning

We assessed learning using two measures across the pre-exposure and exposure phases.

On channel trials we measured percent adaptation as the slope of the regression of the time course of the force that participants produced into the channel wall against the ideal force profile that would fully compensate for the field. To do this we extracted a 400 ms (or the maximum available) window of data centred on the time of peak velocity and calculated the force generated against the channel. We used the velocity along the channel to predict the force the vBOT would have applied on an exposure trial. We performed regression (with zero intercept) on these times series and expressed the slope as a percentage (slope of 1 = 100%), termed adaptation, as in Chapters 2 and 3. In addition, on null and exposure trials, we calculated the maximum perpendicular error (MPE) as in previous chapters.

We averaged adaptation and MPE separately for each participant within a block. We evaluated the difference for each measure between the average of the first 5 short exposure blocks and the final 3 long exposure blocks using Wilcoxon signed rank tests across subjects, due to violations of normality on some measures.

To display hand paths (Figure 4.2), we extracted position data from when the hand left the central target until it entered the secondary target position. Each path was then linearly interpolated (x and y separately) so as to sample 1000 points equally spaced in time. For each subject, we generated a mean path by averaging the sample paths over trials of interest.

To generate a path for a group we calculated the average (and s.e.) of the subjects’ paths and plot the mean with shading showing ± s.e.

Assessments of decay

To examine how contextual de-adaptation depends on planning, execution and time, we evaluated de-adaptation as the difference in adaptation for the left and right lead-in channel trials before and after the intervening trials of each type. That is, because one channel trial for each lead-in start location was performed both at the start and end of every probe block, a single probe block provided two pre-post measures of how de-adaptation depended on each context (same or different), for the particular type of intervening trials assessed (planning,

execution, or time). For example, a set of channel trials originating at the left start location provided a measure of de-adaptation for the ‘same context, execution’ condition when the pre-post adaptation levels for the left start location were evaluated, as well as a measure of de-adaptation for the ‘different context, execution’ condition when the pre-post right start location was evaluated. We averaged the pre-post difference in adaptation across all included probe block repetitions for each participant and probe type, and then averaged across participants.

As some measures of de-adaptation were slightly non-normal, we performed Wilcoxon signed rank tests to evaluate de-adaptation within each condition. To compare across conditions we first performed a Friedman test of de-adaptation (pre-post difference) across all probe conditions (6 levels). We then investigated de-adaptation differences post-hoc for specific pairs of conditions using Wilcoxon signed rank tests.

We took the decay duration of a probe block to be the time from the end of the first two channel trials, to the start of the final two channels, and compared the decay duration of each probe block type (execution, planning-only or wait) collapsed across both contexts using a Friedman test with 3 levels (one for each probe type).

Exclusion criteria

To avoid conflating de-adaptation from executing overshoot mistrials with that of planning different movements, we analysed only probe blocks in which no planning-only trials overshot. This resulted in planning-only probe block data from 15 out of 20 subjects remaining for analysis in the final dataset. As our primary aim was to compare de-adaptation under planning with that of execution, we therefore excluded all data from the 5 subjects who did not manage to complete any planning-only probe blocks without overshoot (total of 48 out of 120 planning-only block repetitions included for the remaining 15 subjects). Of the remaining 15 participants, they overshot the central target at least once on an average of 60.0

± 5.6% out of the 8 planning only probe blocks (range 25.0 - 87.5%) and these probes blocks were excluded. Across subjects, the overshoot over the 8 planning-only blocks occurred at an approximately constant rate (mean of 10% of planning-only trials per block) suggesting that subjects continued to plan movements to the secondary target across the experiment.

Non-parametric tests were performed in cases where Shapiro-Wilk tests yielded evidence against the data being normally distributed. Statistical significance was generally considered

at p < 0.05. When performing posthoc comparisons of de-adaptation between conditions, we used the Bonferroni-corrected α threshold of p < 0.0055 to take into account all 3 ³₂
possible pairwise comparisons. That is, pairwise comparisons for the two different con-texts within a probe type, and pairwise comparisons across probe types within a single context. Similarly, when performing pairwise comparisons of GLM decay rates, we used the Bonferroni-corrected significance threshold of p < 0.0083 (6 comparisons), as there was just a single decay rate for time.